Cloudflare Recovers 100 TB of RAM by Optimizing Pingora’s Consistent‑Hashing Algorithm

Cloudflare’s Performance team identified that its Pingora Backend Router (PBR) was using more memory than expected because the pingora‑ketama library’s consistent‑hashing implementation allocated large hash ranges per server. By revising the algorithm to use multiple hashes per server and applying Rust‑based optimizations, the team reduced the memory footprint of the service. The change freed more than 100 TB of RAM across Cloudflare’s global infrastructure, on top of the 100 TB the DNS team shed last month. The improvement demonstrates how small, mathematically grounded tweaks can yield large savings at scale, allowing Cloudflare to run more services on the same hardware and maintain equitable resource sharing among teams.